CRM hygiene without a data-entry hour
Lesson 3 of 6 · Run it like a shop
1 · Learn the move · Constraints + negative instructions
A CRM rots one shortcut at a time: a note that says 'call back' with no date, a second record for the same owner under a nickname, a request to stop contact typed into a comment where no filter will ever find it. Using a model to turn raw notes into clean fields saves the hour, and it also invites the worst kind of error: fields filled with something plausible. The constraints do the work. Every field comes from a note or reads [NOT IN NOTE]. No field is inferred from another. Two records are never merged by the model, only flagged as possible duplicates with the evidence side by side. And anything that sounds like a request to stop contact is copied word for word into its own field and never softened, cleared or reinterpreted, because that flag is the record of what a person asked you, and the rules for honouring it belong to counsel.
Turn these raw call, mail and text notes into rows with exactly these fields: [fields]. A field comes only from a note, or reads [NOT IN NOTE]. Never merge records; flag possible duplicates with both rows shown. Any stop-contact language goes into DNC_REQUEST word for word, and you never clear an existing flag.
2 · Your turn. You write the prompt
You have a week of scribbled call notes, a stack of returned-mail slips and a text thread export, all about parcels in Vessey County, and the CRM has not been touched since Monday. Two notes might be the same family. One owner said something on the phone that may have been a request to stop calling. Write the prompt that cleans it without inventing a field or losing that request.
Write the prompt for this scenario as given, or swap in a real one from your own work. The score works the same either way.
Remember: the AI sees only your prompt, not this page. If the situation isn't in your prompt, it doesn't exist.
Optional. These shape the output when you run your prompt below, not your score.